who su ers from the covid-19 shocks? labor market ... · besides the three data sources for labor...
TRANSCRIPT
-
Who Suffers from the COVID-19 Shocks?
Labor Market Heterogeneity and Welfare Consequences in Japan ∗
Shinnosuke Kikuchi† Sagiri Kitao‡ Minamo Mikoshiba§
November 13, 2020
Abstract
Effects of the COVID-19 shocks in the Japanese labor market vary across workers
of different age groups, genders, employment types, education levels, occupations,
and industries. We document heterogeneous changes in employment and earnings in
response to the COVID-19 shocks, observed in various data sources during the initial
months after the onset of the pandemic in Japan. We then feed these shocks into
a life-cycle model of heterogeneous agents to quantify welfare consequences of the
COVID-19 shocks. In each dimension of the heterogeneity, the shocks are amplified
for those who earned less prior to the crisis. Contingent workers are hit harder
than regular workers, younger workers than older workers, females than males, and
workers engaged in social and non-flexible jobs than those in ordinary and flexible
jobs. The most severely hurt by the COVID-19 shocks has been a group of female,
contingent, low-skilled workers, engaged in social and non-flexible jobs and without
a spouse of a different group.
Keywords: COVID-19, Japan, labor market, welfare effect, life-cycle model,
inequality
JEL Classification: E21, E24, J31
∗We thank Anton Braun, Daiji Kawaguchi, Masayuki Morikawa, Makoto Nirei, Koki Okumura, Etsuro
Shioji, Kazunari Tanabe, Satoshi Tanaka, Hajime Tomura, Yuichiro Waki, and participants of the seminar
at Waseda University for helpful comments. We acknowledge financial support from the Center for
Advanced Research in Finance (CARF) at the University of Tokyo.†Massachusetts Institute of Technology, Email: [email protected].‡University of Tokyo, Email: [email protected].§University of Tokyo, Email: [email protected]
-
1 Introduction
The COVID-19 pandemic has brought significant shocks to the labor markets all over
the world, and Japan is no exception. While Japan did not see a sharp increase in
unemployment rate immediately after the onset of crisis, which stood at 2.6% in April
2020, compared to other countries such as 14.9% in the United States in April 2020,
shocks in the labor market are spread highly unequally across workers.1,2 In this paper,
we first document heterogeneous responses in employment and earnings to the COVID-19
shocks observed during the initial months after the onset of the crisis in Japan. We then
feed these shocks in the labor market into a life-cycle model of heterogeneous agents to
quantify welfare consequences of the COVID-19 shocks.
Despite the relatively small change in the overall unemployment rate, we find that
negative effects of the COVID-19 shocks significantly differ across individual workers,
in various dimensions including age group, gender, employment type, education level,
occupation, and industry. Moreover, in each dimension, the shock is larger for those who
earned less prior to outbreak of the pandemic, amplifying inequality in the labor market
across multiple dimensions.
To quantify welfare effects from the COVID-19 shocks, we build a life-cycle model and
let heterogeneous individuals face unexpected changes to their earnings and employment,
as observed in the data, and have them re-optimize in response to the shocks. We evaluate
welfare effects on different types of individuals in terms of consumption equivalent varia-
tion that would make them as better off as before in the economy without the COVID-19
shocks.
Our findings can be summarized as follows. First, contingent workers suffer signifi-
cantly, up to more than three times as much as regular workers in terms of our welfare
measure. They are more severely hurt in both employment and wages than regular work-
ers, and we find that employment type is one of the most critical dimensions that divides
the fate of individuals in the labor market after the crisis. Second, we also find that
younger generations suffer more than older generations. Third, female workers fare worse
than males. The difference is mainly due to the fact that the share of contingent workers
is larger for females, but also because females are more concentrated in jobs that are more
severely affected by the COVID-19 shocks. Forth, workers in social sectors and/or non-
flexible occupations suffer more. The COVID-19 crisis differs from past recessions such as
1The Japanese unemployment rate is from the Labor Force Survey (LFS) of the Ministry of Internal
Affairs and Communications (MIC). The U.S. unemployment rate is from the Labor Force Statistics of
the Current Population Survey (CPS). The U.S. unemployment rate peaked in April and declined to
13.3% and 11.1% in May and June, respectively. The Japanese unemployment rate has been relatively
stable during the period with 2.9% in May and 2.8% in June, respectively.2Kikuchi et al. (2020) discuss heterogeneity of potential vulnerability of workers to the COVID-19
shocks using data prior to the crisis.
2
-
the financial crisis of 2008 in that it contracts economic activities in sectors that involve
more face-to-face transactions and occupations involving tasks difficult to be completed
remotely from homes or in physical isolation from other people. Kikuchi et al. (2020) dis-
cussed heterogeneous vulnerability across occupations and industries and pointed to risks
of rising inequality, which we confirm has manifested in wage and employment changes
across workers in the data during the first months after the crisis.
We also stress a caution in the interpretation of our quantitative results. As discussed
above, the main focus of our paper is to assess changes in the labor market during the
initial months after the onset of the COVID-19 crisis, which we observed in various official
data, and to quantify welfare implications from these observations. For this purpose, we
build a simple life-cycle model of heterogeneous agents that enables us to focus on the
analysis of these effects in the short-run. There is, however, significant uncertainty about
whether various shocks we currently observe will be short-lived or long-lived and whether
they will be repeated multiple times over years to come. We evaluate welfare effects under
some scenarios about the duration of shocks and our results may need to be re-examined
when more data is available and there is less uncertainty as to the magnitude and the
duration of the pandemic.3
Moreover, there may well be other structural changes in the economy that the COVID-
19 crisis may induce over the medium and long-run. There are also many changes that the
Japanese economy had been going through, including changes in the composition of em-
ployment type and gender-specific involvement in the labor market, aging demographics,
fiscal challenges associated with rising expenditures on the social insurance system. The
COVID-19 crisis may interact with these changes and possibly amplify challenges that
Japan is faced with in some dimensions, or hopefully mitigate them in other dimensions.
Although we acknowledge these topics and potential consequences of the COVID-19 crises
in the medium and long-term as very important and worth exploring, they are not in the
scope of the current analysis, and our model intentionally abstracts from them. Our focus
is on a quantitative evaluation of shocks in the labor market immediately after the crisis
hit the economy, and we do not explicitly discuss or evaluate specific policies.4
Numerous studies have emerged that investigate heterogeneous consequences of the
COVID-19 shocks on individuals and implications for welfare and policies, which include
but are not limited to Acemoglu et al. (2020), Alon et al. (2020a), Glover et al. (2020),
Kaplan et al. (2020b), and Albanesi et al. (2020), just to name a few.5 Our study
3Some papers including Kawaguchi and Murao (2014), Guvenen et al. (2017) and Huckfeldt (2016)
argue that recessions could have lasting scarring effects on a vulnerable group of workers, especially on
the young.4See Ando et al. (2020) for a comprehensive overview of various policies implemented by the Japanese
government in response to the COVID-19 shocks.5Other papers that document and study early responses to the COVID-19 shocks in the U.S. labor
market include Coibion et al. (2020), Gregory et al. (2020) and Kahn et al. (2020).
3
-
complements the literature by documenting facts and analyzing welfare consequences in
Japan.
This paper is also complementary to studies of various economic aspects of the COVID-
19 shocks in Japan. They include Fukui et al. (2020) on the impact of pandemic on job
vacancy postings, Watanabe and Omori (2020) on consumption responses across sectors,
Miyakawa et al. (2020) on firm default, Kawata (2020) on occupational and spatial
mismatch, Kawaguchi et al. (2020) on uncertainty faced by small and medium-sized
firms, and Okubo (2020) on implementation of telework across occupations.
The rest of the paper is organized as follows. Section 2 provides an overview of
economic shocks triggered by the COVID-19 shocks observed in the early data and lays out
facts that our model analysis in the following sections is focused on. Section 3 presents our
dynamic life-cycle model and section 4 discusses parametrization of the model. Numerical
results are discussed in section 5 and section 6 concludes. The appendices provide more
details about the data sources and discusses our computation methods.
2 Impact of the COVID-19 Shocks on the Labor Mar-
ket in Japan
This section documents changes in employment and earnings during the COVID-19 crisis.
The data source of our analysis is mainly Labor Force Survey (LFS) data for monthly
employment, and is supplemented by Monthly Labor Survey (MLS) data for monthly
earnings and Employment Status Survey (ESS) data in 2017 for composition of workers
across different categories.
2.1 Data Sources
We provide a brief explanation of the three labor market data sources: LFS, MLS, and
ESS below. Detailed description of these data sets is provided in appendix A.
Labor Force Survey (LFS): The LFS is a monthly cross-sectional household survey
conducted by the Ministry of Internal Affairs and Communications (MIC). It covers ap-
proximately 40 thousand households across the nation and collects detailed information
about the employment status of household members. We use publicly available tabulated
data to compute employment by age, gender, employment type, industry, and occupation.
Monthly Labor Survey (MLS): The MLS is a monthly cross-sectional monthly
survey conducted by the Ministry of Health, Labour and Welfare (MHLW), which covers
approximately 33 thousand establishments and their employees from the private and pub-
4
-
lic sectors. We use publicly available tabulated data to compute earnings by employment
type and industry.
Employment Status Survey (ESS): The ESS is a cross-sectional household survey
conducted every five years by the MIC. For our research purpose, we use the latest data
collected in October 2017. It is one of the most comprehensive surveys on employment cir-
cumstances in the nation. It covers approximately 490 thousand households and provides
detailed information about the demographic characteristics of households, employment
and unemployment situations, and descriptions of current jobs held by household mem-
bers. We use the “order-made” summarization system to compute joint distribution of
workers and earnings prior to the crisis, across age groups, genders, education levels,
employment types, occupations, and industries.6
Besides the three data sources for labor market statistics, we also use the Family
Income and Expenditure Survey (FIES) data for changes in consumption level and allo-
cations. More details about the data sources are provided in appendix A.
2.2 Classification of Workers
We briefly explain below how we classify workers according to three different dimensions:
employment type, industry and occupation. More details about the classifications in each
of the data sources are given in appendix A.
Employment-Type Categories: Employment in the Japanese labor market is char-
acterized by a distinction in employment type: regular or contingent employment. How
they are termed in the Japanese language differs depending on situations and data source.
In the ESS, for example, regular employment includes executives of companies and staff
members who are termed regular (seiki) employees. Contingent (hiseiki) employment
includes part-time workers, albeit (temporary workers), dispatched workers, contract em-
ployees and others. Contingent workers are sometimes termed irregular or non-regular
workers as well.7
The distinction is different from that between full-time and part-time workers in other
countries. Contingent workers may well work for the same number of hours as regular
workers but they tend to receive lower wages, fewer fringe benefits, and much less job
security than regular workers. As documented in papers such as İmrohoroğlu et al. (2016)
and Kitao and Mikoshiba (2020), earnings of contingent workers are much lower among
6The ESS data is based on statistical products provided by the Statistics Center, an independent
administrative agency based on the Statistics Act, as a tailor-made tabulation of the 2017 ESS compiled
by the MIC.7How workers are divided into the two employment types in each database we used is explained in
appendix A.
5
-
both males and females. Females have a higher fraction of contingent workers than males
and so do less educated workers than those with higher education. Moreover and most
importantly, contingent workers are subject to more frequent employment adjustment and
job instability, as shown in empirical studies including Esteban-Pretel et al. (2011) and
Yokoyama et al. (2019). In the analysis below, we include employment status as one of
the key dimensions of heterogeneity across workers in evaluating effects of the COVID-19
crisis.
Sectoral Categories: The second dimension along which we classify workers is in-
dustry. The COVID-19 crisis invokes urgent need to suppress economic activities that
require personal interactions, in order to control the infections. Workers in industries
that produce goods and services that involve a large amount of personal and face-to-face
interactions are affected more severely by the COVID-19 shocks. This is in a unique
feature of the COVID-19 crisis, leading to a group of damaged industries that differ from
a typical set of industries that are more likely to be hit by regular recessions.
Following Kaplan et al. (2020b), we group industries into two sectoral categories:
ordinary and social.8 Based on the distribution of workers across sectors in the ESS,
48% of total employment is classified into the ordinary sector, and the remaining 52% is
classified into the social sector, prior to the COVID-19 shocks.
• Ordinary Sector: agriculture, forestry and fisheries; mining, quarrying of stone andgravel; electricity, gas, heat supply and water; construction; manufacturing; whole-
sale; transport and postal activities except for railway, road passenger and air trans-
port; postal service; information and communications; finance and insurance; real
estate, goods rental and leasing.
• Social Sector: retail trade; railway, road passenger and air transport; educationand learning support; medical, health care and welfare; living-related, personal and
amusement services; accommodations, eating and drinking services; scientific re-
search, professional and technical services; cooperate associations, n.e.c.; services,
n.e.c.; government.
Note that not all data sources provide sector information of the same accuracy, and we
use a broader classification for the MLS. Also, we use a slightly different categorization
for the expenditure data from the FIES. For more details, see sections A.2 and A.4,
respectively.
8We use industrial categories defined in the Japan Standard Industrial Classification (JSIC), as revised
in 2013. Although we follow the classification of Kaplan et al. (2020b), their classification according to
the NAICS does not match exactly our classification in the ESS based on the JSIC. See appendix A for
more details of our industry classification. Please note that Kaplan et al. (2020b, 2020a) use the term
“regular” industries to denote what we call “ordinary” industries here. “Regular” is used to represent
one of two employment types in this paper.
6
-
Occupational Categories: Finally, we classify workers by occupations and group
them into two occupational categories, flexible and non-flexible occupations, based on the
fraction of workers in each occupation who are likely to work remotely and less affected
by difficulty in commuting to and working in their regular workplace.9 Following Mongey
et al. (2020), we construct measures of the fraction of flexible-type workers in each
occupation. Figure 1 shows the result. We then classify occupations as flexible if the
measure is larger than 0.75. As a result, 60% of total employment is classified into
flexible occupation, and the remaining 40% is classified into non-flexible occupation.
0 .25 .5 .75 1Fraction of work-from-home
Construction and miningManufacturing process
SecurityCarrying, cleaning, packaging
Transport and machine operationService
Agriculture, forestry, fisherySales
Professional and engineringClerical
Administrative and management
WFH Not WFH
Figure 1: Work-from-home Measures: JSOC
Note: This figure shows the fraction of workers who are able to work from home in each occupation. To
compute the measure, we follow Mongey et al. (2020) and convert the Standard Occupational Classifi-
cation (SOC) to the Japan Standard Occupational Classification (JSOC).
• Flexible Occupation: administrative and management; clerical workers; professionaland engineering workers; sales workers.
• Non-flexible Occupation: agriculture, forestry and fishery workers; service workers;transport and machine operation workers; carrying, clearing, packaging and related
workers; security workers; manufacturing process workers; construction and mining
workers.
9We use occupational categories defined in the Japan Standard Occupational Classification (JSOC),
as revised in December 2009.
7
-
2.3 Changes in Employment
This section documents changes in employment in Japan during the COVID-19 crisis.
The data source is LFS data for most of the analysis, and ESS data for compositional
analysis.10
By Employment Type, Sector and Occupation: Figure 2a shows the number
of employed by employment type (regular and contingent). We normalize to 100 the
level of employment for each type in January 2020. While regular workers’ employment
declined by around 1% in April, May, and June compared to January, contingent workers’
employment declined more sharply by around 4%. An observation that employment of
contingent workers responds more to shocks is consistent with previous crisis episodes in
Japan where contingent workers have been more vulnerable to business cycle shocks, as
documented by Yokoyama et al. (2019).
Figure 2b shows the number of employed according to the sectoral and occupational
categories defined above. The number of workers in the social sector and non-flexible
occupations declined the most, by more than 5% from January to April 2020, and it
remains low until June. The difference across sectors and occupations highlights the
importance of the feasibility of completing work from home, as emphasized by Dingel
and Neiman (2020) in the case of the US labor market and Fukui et al. (2020) based on
changes in the pattern of job vacancy postings in Japan after the COVID-19 shocks.
10Note that the monthly data series presented in sections 2.3 are raw data and not seasonally adjusted
and changes include potentially seasonally factors. See appendix D for seasonally adjusted versions of
the same figures.
8
-
9510
010
5#
of th
e em
ploy
ed (
Jan
2020
=100
)
1 2 3 4 5 6Month
RegularContingent
(a) By Employment Type
9510
010
5#
of th
e em
ploy
ed (J
an 2
020=
100)
1 2 3 4 5 6Month
Ordinary and FlexOrdinary and Non-FlexSocial and FlexSocial and Non-Flex
(b) By Sector and Occupation
Figure 2: Changes in Employment (Jan. 2020 = 100)
Note: Figure 2a shows the number of employed by employment type in each month between January
and June 2020, based on samples of workers aged 25 to 64. Figure 2b shows the number of employed by
sector and occupation categories. The samples include workers aged 15 to 64, differently from Figure 2a,
since data by more granular categories is not available from the publicly available aggregate data. For
the same reason, samples in Figure 2b include not only regular and contingent workers but also other
types of workers such as the self-employed. In both figures, the values in January 2020 are normalized to
100, and series are not seasonally adjusted. The data is from Labor Force Survey (LFS) by the Ministry
of Internal Affairs and Communications (MIC).
By gender: Figure 3 shows changes in the number of employed by gender, where the
level in January 2020 is normalized to 100. While both males’ and females’ employment
declined since February 2020, the decline is larger for females. This is similar to what
occurred in the U.S. where female workers were hit harder by the COVID-19 shocks than
male workers, as emphasized by Alon et al. (2020a).
9
-
9798
9910
010
1#
of th
e em
ploy
ed (J
an 2
020=
100)
by
gend
er
1 2 3 4 5 6Month
Male Female
Figure 3: Changes in Employment by Gender (Jan. 2020 = 100)
Note: Figure 3 shows the number of employed by gender in each month between January and June 2020.
We restrict samples to workers aged 25 to 64. The values in January 2020 are normalized to 100, and
series are not seasonally adjusted. The data is from Labor Force Survey (LFS) by the Ministry of Internal
Affairs and Communications (MIC).
Why have female workers suffered more from the COVID-19 shocks? Figure 4 shows
the characterization of workers by gender based on the ESS data prior to the COVID-19
crisis. Figure 4a displays the share of contingent workers out of total employment by
gender. While the share of contingent workers is less than 10% for males, more than 50%
of female workers work a contingent job. This difference partially contributes to larger
decline for female employment, since contingent workers are subject to more employment
adjustment during economic downturns as discussed above, and in fact, there was a larger
decline in employment among contingent workers as we show below.
Figure 4b shows the share of workers in the social sector out of total employment by
gender. Again, female workers are more concentrated in the social sector (69%) than male
workers (39%). Figure 4c shows the share of workers in non-flexible occupations out of
total employment by gender. In contrast to employment type and sector, male workers
appear to be more vulnerable in terms of the non-flexibility of the work arrangement,
though the difference is relatively small.11 Figure 4d, however, which shows the joint
distribution of employment across sectors and occupations, reveals that the share of the
most vulnerable workers engaged in social and non-flexible jobs is higher for females than
males. The share of the least vulnerable workers in ordinary and flexible jobs is larger for
11The share of non-flexible occupations is 46% for males and 34% for females.
10
-
males than females as well.
0 20 40 60 80 100
Female
Male
Regular Contingent
(a) By Employment Type
0 20 40 60 80 100
Female
Male
Ordinary Social
(b) By Sector
0 20 40 60 80 100
Female
Male
Flexible Non-flexible
(c) By Occupation
0 20 40 60 80 100
Female
Male
Ordinary and Flex Ordinary and Non-FlexSocial and Flex Social and Non-Flex
(d) By Sector-Occupation
Figure 4: Share of Each Characteristics by Gender
Note: Figure 4 shows the employment share for each characteristic by gender. We restrict samples to
workers aged between 30 and 59 because the data is available only for 10-age bin. The data is from Em-
ployment Status Survey (ESS) conducted in 2017 by the Ministry of Internal Affairs and Communications
(MIC).
By Age Group: Figures 5a and 5b show the number of employed by age for regular
workers and contingent workers, separately. We normalize the level in January 2020 to
100. For regular workers, changes during the first six months of the year are modest.
For contingent workers, the decline by April 2020 is much larger in the range of 4 to 5%
relative to the level in January 2020. Across age groups, changes from January 2020 to
April 2020 are similar, but the decline from the first quarter to April and May of 2020
is larger for younger cohorts. Nonetheless, recovery is faster for younger cohorts in June
11
-
as well. We discuss this heterogeneity in employment across age groups and employment
types in more details in section 5.2.90
9510
010
5#
of th
e em
ploy
ed (
Jan
2020
=100
)
1 2 3 4 5 6Month
Age 30/39 Age 40/49Age 50/59
(a) Regular Workers
9095
100
105
# of
the
empl
oyed
(Ja
n 20
20=1
00)
1 2 3 4 5 6Month
Age 30/39 Age 40/49Age 50/59
(b) Contingent Workers
Figure 5: Changes in Employment by Age Group (Jan. 2020 = 100)
Note: Figure 5a shows the number employed by age for regular worker in each month between January
and June 2020. Figure 5b shows the number of employed by age for contingent workers during the same
period. The values in January 2020 are normalized to 100. Samples are restricted to workers aged 25 to
64. Series are not seasonally adjusted. The data is from Labor Force Survey (LFS) by the Ministry of
Internal Affairs and Communications (MIC).
2.4 Changes in Earnings
This section documents changes in earnings in Japan during the COVID-19 crisis, based
on the MLS data. Figure 6 shows year-on-year changes in earnings in the ordinary and
social sectors for regular workers and contingent workers, separately. Note that, in MLS,
we use data for part-time workers as that for contingent workers due to data availability.12
As shown in Figure 6a, earnings of regular workers barely changed during the first
quarter of 2020 compared to the same months of the previous year. The average earnings
in both sectors declined in the second quarter of 2020, by 1 to 2% in the social sector
compared to the same periods in 2019, and by 1 to 4% in the ordinary sector.
6b shows year-on-year changes in earnings for contingent workers in the first half of
2020 with significant differences in the changes across sectors. For workers in the social
sector, earnings declined by 4 to 5% in April and May while those in ordinary sectors
12See section A for more details about the classification of employment types and 5.4.2 for further
discussion on the difference in the classifications between the MLS and the LFS (or ESS) and its effects
on the calibrated wage shocks.
12
-
experienced a relatively modest decline. There is a sharp rise in earnings of contingent
workers in the social sector in June and a mild increase of those in the ordinary sector.13-8
-6-4
-20
24
Cha
nges
in to
tal a
vera
ge e
arni
ngs
(YO
Y in
%)
1 2 3 4 5 6Month
Ordinary Social
(a) Regular Workers
-8-6
-4-2
02
4C
hang
es in
tota
l ave
rage
ear
ning
s (Y
OY
in %
)
1 2 3 4 5 6Month
Ordinary Social
(b) Contingent Workers
Figure 6: Changes in Earnings by Sector (in YOY change, %)
Note: Figure 6a and 6b show changes in fixed earnings by sector for regular and contingent workers (we
use part-time workers), respectively. The values are in year-on-year change by monthly frequency, that
is, they compare changes in earnings from the same month in the previous year. Series are not seasonally
adjusted. The data is from the Monthly Labor Survey (MLS) by the Ministry of Health, Labour and
Welfare (MHLW).
3 Model
Demographics: At age j = 1, individuals enter the economy with initial assets de-
noted as a1. Individuals face probability sj of surviving from age j − 1 to j. Sj denotesunconditional survival probability that an individual lives up to age j. We assume that
they retire at the age of j = JR and live up to the maximum age of j = J . The de-
ceased will be replaced by the newborn. Population is assumed to be constant and age
distribution is stationary.
Endowment and Earnings: Individuals are born with gender g = {M,F}, male orfemale, and a skill type s = {H,L}, high or low. Upon entering the labor market, theyare also assigned to an employment type e = {R,C}, regular or contingent, an occupationo = {o1, o2}, and sector d = {d1, d2}.
13The rise in the number of contingent workers in the social sector in June is driven by a recovery in
earnings of retail, medical care and welfare and educational services industry, which offsets a continued
decline in accommodation, eating and drinking services industries.
13
-
The two occupation types, o1 and o2, are associated with different levels of work
flexibility, i.e. whether the job can be done remotely from home or not. The two sectors,
d = {d1, d2}, produce different types of goods and services. Sector d1 produces ordinarygoods while sector d2 produces social goods, which are more immune to infection risk in
terms of consumption.
We let x = {j, g, s, e, o, d} denote a state vector of each individual. We denote by µxthe population share of individuals in state x, that is, age j, gender g, skill s, employment
type e, occupation o, and sector d. Each individual’s efficiency units of labor depend on
the state vector x and are denoted as ηx, which varies over a life-cycle and approximates
human capital that grows in age for each type of workers.
Earnings of an individual in state x at time t are given by
yx,t = λx,tηxwt.
λx,t summarizes shocks that affect earnings of type-x individuals at time t, which will be
discussed in detail in section 5.2. wt denotes the market wage per efficiency unit of labor.
Preferences: Individuals derive utility from consumption of two types of goods, c1
and c2, representing ordinary and social goods, respectively. We assume a period utility
function:
U(c1, c2) = ξt
[cγt1 c
1−γt2
]1−σ1− σ
, (1)
where ξt represents an intertemporal preference shifter that affects marginal utility from
consumption in each period. It is a weight on utility from consumption at time t relative
to other times and may change with the arrival of the COVID-19 shocks, but it is assumed
to be constant in normal times.
γt is a preference weight on ordinary goods, which, similarly to ξt, is constant in
normal times, but may vary upon the arrival of the COVID-19. σ represents risk aversion.
Individuals discount future utility at constant rate β.
There are no bequest motives and assets at+1 left by the deceased are collected and
transferred to all surviving individuals as accidental bequests, denoted as bt, which satisfies
the following equation.
bt =
∑x at+1(x)(1− sj+1)µx∑
x µx(2)
Government: The government operates a social security program, which provides
a pension benefit pt to each retiree. Individuals are taxed on their consumption, labor
income and capital income at proportional rates, τc,t, τl,t, and τa,t, respectively. We
assume that the government budget is balanced each period and let a lump-sum transfer
τls,t absorb an imbalance from the period budget constraint (3).
14
-
∑x
[τc,t(c1,t(x) + c2,t(x)) + τa,trt(at(x) + bt) + τl,tλx,tηxwt]µx =∑x|j≥jR
ptµx +∑x
τls,tµx
(3)
Life-cycle Problem: The intertemporal preference ordering of an individual of type
x born at time t is given by:
U({c1,t+j−1, c2,t+j−1}Jj=1) =J∑j=1
βj−1Sjξt+j−1
[cγt+j−11,t+j−1c
1−γt+j−12,t+j−1
]1−σ1− σ
subject to:
(1 + τc,t)(c1,t + c2,t) + at+1 = (1− τl,t)λx,tηxwt +Rt(at + bt) + τls,t for j < jR
(1 + τc,t)(c1,t + c2,t) + at+1 = pt +Rt(at + bt) + τls,t for j ≥ jR
where Rt = 1 + (1 − τa,t)rt denotes net-of-tax gross interest rate at time t. We assumethat the relative price of ordinary and social goods is constant and normalized to 1.
Initial Economy and Transition Dynamics The initial economy is stationary and
characterized by demographics, {sj}Jj=1 and µx, type-specific labor productivity, ηx, aset of fiscal variables, {τc, τl, τa, p}, factor prices, {r, w}, where individuals choose theoptimal path of consumption and assets {c1, c2, a′} at each age j. In equilibrium a lump-sum tax, τls, balances the government budget (3) and the accidental bequest, b, satisfies
the condition (2).
At time 1, we assume that individuals are hit by wage and employment shocks summa-
rized in λx,t, which we will fully characterize in section 5.2, as well as by preference shocks,
ξt and γt. Given the new paths of earnings and preferences, individuals re-optimize and
choose a new path of consumption and assets. We let τls,t adjust to balance the gov-
ernment budget to satisfy (3) in each period as well bequests bt to meet the condition
(2).
4 Calibration
This section describes parametrization of the economy presented above. The model fre-
quency is quarterly. The initial economy approximates the Japanese economy prior to
the onset of the COVID-19 shocks. We compute the transition dynamics starting in the
first quarter of 2020, which corresponds to our initial economy. Parametrization of the
initial economy is explained in this section and summarized in Table 1. The shocks that
characterize the COVID-19 crisis are discussed in section 5.2.
15
-
4.1 Demographics
Individuals of the model enter the economy and start working at the age of 25, and they
may live up to the maximum age of 100 years subject to age-specific survival probabilities
sj. The retirement age jR is set at 65 years old. We calibrate the probabilities based on
the estimates of the National Institute of Population and Social Security Research (IPSS)
for the year 2020. We abstract from population growth and age distribution is stationary.
4.2 Preferences
The risk aversion parameter, σ, in the utility function (1) is set to 2.0. The parameter
γ in the initial economy represents a weight on ordinary goods relative to social goods
and it is set at 0.789 so the model matches the ratio of consumption expenditures of the
two types of goods, based on the Family Income and Expenditure Share (FIES) from the
Ministry of Internal Affairs and Communications (MIC). The parameter ξ that represents
an intertemporal weight on consumption is set at 1 in the initial economy. In section 5.4,
we simulate time-varying preference weights to approximate consumption data observed
during the initial months of the COVID-19 crisis.
The subjective discount factor β is set at 1.0014 (or 1.0054 on an annual basis) to
match the average growth of consumption between ages 25 and 50 as observed in the
FIES data estimated in İmrohoroğlu et al. (2019).
4.3 Endowment and Human Capital
Each individual is endowed with a unit of time and supplies labor inelastically until
they reach the retirement age jR. The labor productivity ηj,g,s,e,o,d, which represents
human capital of an individual worker and evolves over a life-cycle, is calibrated with the
ESS data. Details about the categorization of individual workers into employment type,
education level, industry and occupation are provided in appendix A.
We assume that the type of individual worker is determined upon entry to the labor
market and fixed throughout their life-cycle. The share of each type is based on the
distribution from the ESS data, and we take the average share of types among individuals
aged between 30 and 59.
4.4 Government and Other Parameters
The pay-as-you-go social security program provides pension benefits p to each retiree. We
assume that benefits are set to 30% of average earnings in the initial economy, based on
the estimated replacement rate of social security benefits by the OECD.14
14OECD Pension at a Glance, 2020.
16
-
The consumption tax rate, τc, is set to 10%. Labor and capital income tax rates,
τl and τa, are set to 13% and 20%, respectively, following İmrohoroğlu et al. (2019).
The lump-sum transfer τls is determined in equilibrium to absorb an imbalance from the
government budget and is set to 7.3% of average earnings in the initial economy.
We set the interest rate at 2%, which is in the range of estimated returns to household
saving, such as Aoki et al. (2016). Wage rate is normalized so that the average earnings
in the initial economy is 1.
Table 1: Parameters of the Model: Initial Economy
Parameter Description Value
Demographics
JR Retirement age 65 years
J Maximum age 100 years
sj Survival probability IPSS data
µj,g,s,e,o,d Population share ESS data
Preference
β Subjective discount factor 1.0054 (annual)
σ Risk aversion parameter 2.0
γ Expenditure share on ordinary goods 0.789 (FIES)
ξ Intertemporal weight 1 (before shock)
Human Capital
ηj,g,s,e,o,d Life-cycle human capital ESS data
λ Shocks to earnings 1 (before shock)
Government
τc Consumption tax rate 10%
τl Labor income tax rate 13%
τa Capital income tax rate 20%
τls Lump-sum tax/transfer 7.3% of avg. earn
p Social security benefit 30% of avg. earn
Other Parameters
r Interest rate 2%
w Wage rate Normalization
5 Numerical Results
5.1 Baseline Model: Initial Economy
Figure 7 shows the earnings profile based on ESS data as discussed in section 4, for
selected types of workers. The left panel shows average earnings of all workers at each
17
-
age, normalized to the average earnings of all workers. It exhibits a hump-shaped profile,
where earnings rise monotonically after the entry and peak at around age 55, when they
start to decline. The right panel shows profiles for each gender and employment type and
highlights a stark difference in earnings by individual characteristics.
(a) By Age (b) By Age, Gender and Emp. Type
Figure 7: Earnings in the Initial Economy (in model units; average earnings=1)
Solving the model described above, we obtain consumption and asset profiles of indi-
viduals averaged for each age, as shown in Figure 8.15
(a) Consumption by Age (b) Assets by Age
Figure 8: Consumption and Assets in the Initial Economy (in model units; average earn-
ings=1)
15Note that assets are expressed in terms of average annual earnings, with an adjustment for quarterly
frequency of the model.
18
-
5.2 The COVID-19 Shocks
We will next discuss the COVID-19 shocks that are introduced in the initial economy
described above, before we study how they affect welfare of heterogeneous individuals in
the model economy in section 5.3. This section revisits the data description presented
in section 2 and explains how we process them as shocks that we feed into our model.
We will decompose shocks into five; three associated with wage and employment shocks
and two associated with preferences. Our main focus will be the first three. Table 2
summarizes five different types of shocks that we consider in the simulations.
Wage and Employment Shocks: Earnings of an individual in state x are hit by
wage and employment shocks, summarized in λx,t ≡ ωe,d,tφo,d,tνj,e,t. This decompositioncaptures shocks to wages, ωe,d,t, and to employment, φo,d,t and νj,e,t.
Wage shocks, we,d,t, are specific to the industry and vary by employment type, and
they are measured as a change in earnings between the first and the second quarters of
2020, using the MLS data.16 The shocks vary across the combination of employment
type and industry, (e, d) = (1,1), (1,2), (2,1), (2,2), independently of other states of an
individual, and are set to {w1,1, w1,2, w2,1, w2,2} = {0.9840, 0.9905, 0.9748, 0.9757} based onthe quarterly change in the data. Workers with contingent employment type experience
a wage decline of 2.5% in the ordinary sector and 2.4% in the social sector, while the
change is relatively small for those engaged in a regular job.
Employment shocks consist of two parts, employment type shock, νj,e,t, and occupation-
sector specific shock, φo,d,t. We calculate the employment type shock, νj,e,t, from a change
in the number of employees between the first and the second quarters of 2020, using the
LFS data.17 Changes in employment by employment type vary by age, and we assume
that the shock is age dependent. Figure 9 displays the decline in employment of contin-
gent workers relative to regular workers. Contingent workers experienced a larger decline
in employment across all age groups and Figure 9 shows that employment type shocks hit
younger workers harder than older workers.
The finding that employment of contingent workers is more vulnerable to shocks is in
line with observations during past recession episodes. We also note that it is in general
difficult for firms to adjust the number of regular workers in a few months after the onset
of the crisis and there is possibility that at least part of the difference in the employment
adjustment between regular and contingent workers may be reflecting the speed that
16We use monthly MLS data since January 2013 to June 2020. Before calculating the shocks, we sea-
sonally adjust raw data by converting data from monthly to quarterly frequency. Please see appendices A
and B for detailed data structures and definitions.17We use monthly LFS data since January 2013 to June 2020. Before calculating the shocks, we sea-
sonally adjust raw data by converting data from monthly to quarterly frequency. Please see appendices A
and B for detailed data structures and definitions.
19
-
different types of workers are affected by the crisis. It remains to be seen how employment
of regular workers may be adjusted during the next quarters, especially if the shocks turn
out to be very persistent.18
Figure 9: Employment-type Shocks by Age: Change in Employment of Contingent Work-
ers relative to Regular Workers (Regular=1, 2020Q1 vs 2020Q2)
Note: This graph shows changes in the number of contingent workers relative to regular workers from
age 25 to 65 between the first and second quarter of 2020. Series are seasonally adjusted. The data is
from the Labor Force Survey (LFS) by the Ministry of Internal Affairs and Communications (MIC).
The occupation-sector specific employment shocks, φo,d,t, are computed for each com-
bination of (o, d) = (1,1), (1,2), (2,1), (2,2) and are set at {φ1,1, φ1,2, φ2,1, φ2,2} = {0.9975,1.0021, 0.9902, 0.9508}. Employment of workers engaged in non-flexible occupations inthe social sector is the most severely hurt, falling by 4.9%, while the change is relatively
small for those in ordinary sector, or social but in the flexible occupation.19, 20
18We thank an anonymous referee for pointing out this possibility to us and suggesting that we state
it explicitly.19In computing the decline of employment by occupation and sector, we also use the LFS and ESS data
of MIC. Since the LFS data only observe employment change of all type-(o, d) workers, shocks using only
LFS may be biased by age-composition. Therefore, we use computed employment-age shocks νj,e,t and
the ESS data to isolate shocks associated with industry and occupation in a way that is consistent with
the aggregate changes in employment for each occupation and sector. More details of the computation
are given in appendix B.20Industries that contribute to a rise in the social and flexible group include educational support and
schools.
20
-
Preference Shocks: Preference shocks are captured by share parameter shock, γt,
and intertemporal preference shock, ξt.21 The preference parameters are summarized in
Table 2.
Figure 10 shows the expenditure share for social goods from the FIES data. Until the
first quarter of 2020, the expenditure share of social goods remained stable at 21.1% on
average, and it plummeted by 5.6 percentage points, to 15.5% in the second quarter of
2020. We take this decline in the expenditure share as reflected in the share parameter
shock γt.
We calibrate intertemporal preference shock, ξt, to match the change in total expen-
ditures from the fourth quarter of 2019 to the second quarter of 2020 by using the FIES,
which stands at minus 6.3%. The value of ξt in the first quarter of the shock that generates
a decline in consumption in the observed magnitude is 0.892.
.14
.16
.18
.2.22
2013q1 2014q1 2015q1 2016q1 2017q1 2018q1 2019q1 2020q2Quarter
Figure 10: Expenditure Share of Social Goods
Note: This graph shows the expenditure share of social goods between the first quarter of 2013 and the
second quarter of 2020. The samples are multiple-person households with no restriction of age. Data is
constructed by monthly data from January 2013 to June 2020 by converting to quarterly data. Series
are seasonally adjusted. The data is from the Family Income and Expenditure Survey (FIES) by the
Ministry of Internal Affairs and Communications (MIC).
Table 2 summarizes the shocks observed during the first quarter of the COVID-19
crisis. As we stand, we do not know how long the shocks will remain after the second
quarter of 2020. In the next section, we simulate the transition under some scenarios
21Similarly to wage and employment shocks, we use monthly consumption data, FIES, from January
2013 to June 2020 by converting to quarterly data and seasonally adjusting them. Please see appendices A
and B for detailed data structure and definitions.
21
-
about the duration of the shocks.
Table 2: The COVID-19 Shocks in 2020Q2
Parameter Description Values, source
Wage Shocks
ωe,d,t Wage shock {0.9840, 0.9905, 0.9748, 0.9757}, MLSEmployment Shocks
νj,e,t Employment-age specific shock Figure 9, LFS
φo,d,t Occupation-sector specific shock {0.9975, 1.0021, 0.9902, 0.9508}, LFS and ESSPreference Shocks
γt Share parameter shock 5.6ppt, FIES
ξt Intertemporal preference shock 0.892, FIES
5.3 Transition Dynamics and Welfare Analysis
As discussed in section 5.2, COVID-19 brought sizable shocks to the labor market but the
effects are far from uniform across heterogeneous groups of individuals. We now simulate
the transition of our model economy assuming that individuals in the initial economy are
hit by the shocks at time 1 and make a transition back to normal times over time.
In this section, we first focus on effects of labor market shocks through employment
and wage shocks, explained in section 5.2. In the next section, we will also add shocks
to preferences to account for changes in consumption shares and levels observed in the
data. Our main focus, however, is on effects of heterogeneous labor market shocks on
individuals’ welfare.
As discussed above, it is very difficult, if not entirely impossible, to conjecture how
long the shocks will persist. We assume that the shocks are temporary and disappear
eventually, but will last for multiple periods. In the computation, we let the shocks
diminish at rate ρ each period, with expected duration of 1/ρ.
In the baseline scenario, we assume that shocks last for one year (four quarters) in
expectation and set ρ = 0.25. In section 5.4, we also consider more and less optimistic
scenarios, in which shocks diminish more quickly with expected duration of two quarters,
and more slowly over six quarters, respectively.
Given the size of initial shocks as summarized in Table 2, the average earnings exhibit
a decline of 2.5% in the first quarter of the crisis, which gradually diminishes over the
following quarters, as shown in Figure 11. Note that the decline takes into account changes
in both employment and earnings of individuals.
22
-
Figure 11: Changes in Average Earnings Relative to the Initial Economy (%)
The shocks, however, do not hit individuals equally. Figure 12 shows heterogeneity
in the magnitude of shocks by gender, education level, and employment type under the
baseline scenario where expected duration of shocks is four quarters. They are expressed
as a percentage change in earnings of each type of worker relative to the levels in the
initial economy.
As shown in Figure 12a, females on average experience a 2.9% drop in earnings while
the decline is 2.3% for males. Figures 12b and 12c show an even starker difference in the
decline of earnings across employment types and education levels of workers. Contingent
workers experience a drop of 6.0% on average, while earnings of regular workers decline
by 2.0%. Individuals with less than a college degree experience a sharper decline than
those with a college degree. Note that we do not have any education-specific shock in the
model and the difference comes from different compositions of workers within each group
that are hit by the COVID-19 shocks.
23
-
(a) By Gender (b) By Employment Type
(c) By Education
Figure 12: Changes in Average Earnings Relative to the Initial Economy (%)
We feed these shocks into our model in transition and compute welfare effects on
different types of individuals. We use the initial economy as a basis of comparison and
consider how individuals’ welfare changes once the COVID-19 shocks hit the economy
and they live through the new paths of earnings.
More precisely, we compute welfare of individuals under the initial economy as well as
welfare of all types of individuals in an economy that experiences the COVID-19 shocks at
time 1, which corresponds to the second quarter of 2020. We then compute consumption
equivalent variation, “CEV,” which equals a percentage change in consumption in the
initial economy that would make an individual indifferent between living in the initial
economy versus the economy facing COVID-19 shocks.
In order to account for difference in the expected duration of remaining life, which
varies by individuals of different ages, we compute the present discounted value of con-
sumption adjustment for the rest of an individual’s life, which we call “PV-CEV,” that
24
-
will be needed to make the individual indifferent.22
Tables 3 and 4 show the PV-CEV of different groups of workers relative to average
earnings of each group. Table 3 shows average welfare effects by gender, employment
type and education level. Females on average face a welfare loss equivalent to 2.5% of
their earnings, while the loss is more moderate at 1.8% for males. The table also shows a
significant welfare loss for contingent workers, in a magnitude that corresponds to 4.6%
and 5.1% of earnings for males and females, respectively.
Table 3: Welfare Effects by Gender, Employment Type and Education (aged 25-64, in
PV-CEV)
Emp. type Education
All Regular Cont. High Low
All −1.99 −1.60 −5.00 −1.44 −2.45Male −1.75 −1.64 −4.59 −1.39 −2.13Female −2.53 −1.46 −5.11 −1.62 −2.94
Table 4 shows welfare effects that differ across occupations and industries of individual
workers. Workers in the social sector suffer significantly more from the COVID-19 crisis
than those in the ordinary sector. The negative effect is much larger among those in
non-flexible occupations, conditional on industry. Workers in the ordinary and flexible
jobs experience a small loss of 1.6%, while those in the social and non-flexible jobs suffer
from a large welfare loss of 4.9% relative to their earnings. Within each occupation and
industry, females face a more significant welfare loss than males.
22Denoting the optimal consumptions of an individual in state x before the COVID-19 crisis by
{c∗1,t(x), c∗2,t(x)} and those in the economy hit by the COVID-19 shocks by {c̃1,t(x), c̃2,t(x)}, the CEVfor an individual in state x and aged j at time t is computed as µ(x) that satisfies
J∑k=j
βk−jU(c∗1,t+k−j(x), c∗2,t+k−j(x)) =
J∑k=j
βk−jU(c̃1,t+k−j(x)(1 + µ(x)), c̃2,t+k−j(x)(1 + µ(x))).
The PV-CEV for an individual in state x and aged j at time t is computed as
µ(x) =
J∑k=j
k∏i=j
si/R
/(sj/R) (c̃1,t+k−j(x) + c̃2,t+k−j(x))µ(x).
25
-
Table 4: Welfare Effects by Gender, Industry and Occupation (aged 25-64, in PV-CEV)
Ordinary Social
Flexible Non-flex. Flexible Non-flex.
All −1.57 −2.32 −1.21 −4.86Male −1.42 −2.11 −0.81 −4.20Female −2.09 −3.73 −1.64 −6.00
We now turn our attention to heterogeneity in welfare effects across age groups. Fig-
ure 13 plots the welfare effects by gender and age in 2020. They are expressed in terms
of PV-CEV in units of average earnings of all workers, males, and females, respectively,
in the initial economy. Retirees are not affected directly by the wage shocks but their
welfare declines slightly as we assume that lump-sum transfers are adjusted to make up
for a decline in tax revenues so the government can pay its social security expenditures.
Since individuals approaching the retirement age suffer from a decline in earnings only for
a small number of years, the loss is small relative to younger individuals. Among young
individuals, the magnitude of welfare effects depends on the size of employment shocks
that hit individuals of different age groups and the magnitude of lost earnings relative to
their lifetime average earnings.
For females, the age-wage profile is relatively flat as shown in 7. Moreover, as we saw in
Figure 9, employment of contingent workers is more severely hurt among the young, which
adds to a larger welfare cost for them. The effects of shocks to contingent workers more
clearly manifest among female workers, whose share of contingent workers is much larger
than males. Females are also concentrated in the types of jobs that are more severely hit
by the COVID-19 shocks and it contributes to a larger welfare loss than males. Among
males, the age-wage profile is more hump-shaped than females and peaked at around 55.
Males who are hit by the COVID-19 crisis at around these ages of high earnings suffer
slightly more than younger males, which results in the mildly U-shaped welfare loss of
male workers.
26
-
Figure 13: Welfare Effects by Age and Gender (in PV-CEV)
Figure 14 shows welfare effects by other dimensions of heterogeneity across workers.
As shown in Figure 14a, contingent workers suffer more from the shocks than regular
workers and the difference is larger among younger workers who are hit harder by the
employment type shocks, as discussed in section 5.2. Figure 14b demonstrates that the
low-skilled workers suffer by more than the high-skilled workers across all working ages.
(a) By Employment Type (b) By Education
Figure 14: Welfare Effects by Age, Employment Type and Education (in PV-CEV)
The analysis reveals the fact that negative effects of the COVID-19 crisis in the labor
market have very different implications for people of different age, gender, employment
type, education and job type in terms of industry and occupation. In each dimension, the
shock is larger for those who earn less initially.
Our model captures heterogeneity across workers in many dimensions that turn out to
be critical in evaluating welfare effects the COVID-19 crisis in Japan. There are, however,
other dimensions that are not captured in our model. For example, our model assumes
27
-
full insurance within each group and does not account for within-type heterogeneity in
other dimensions such as wealth, health status, family structure, etc, which presumably
may be important dimensions to analyze once a model is properly extended and calibrated
to data.
In the following section, we run a few additional experiments to consider alternative
scenarios about duration of the COVID-19 shocks, and to introduce preference shocks
to account for changes in consumption level and relative allocation across different types
of goods. We will also consider welfare of some hypothetical households that consist of
different types of individuals.
5.4 Sensitivity Analysis and Alternative Scenarios
5.4.1 Preference Shocks
We now consider shocks to preferences upon outbreak of the COVID-19 crisis. As sum-
marized in section 5.2, there was a sizeable shift in the shares of consumption goods
allocated to ordinary and social goods. The share of the latter was very stable at around
21% before the crisis and plummeted to 15.5% in the second quarter of 2020. At the
same time, when we compare the level between the fourth quarter of 2019 and the second
quarter of 2020, we found the average consumption level also fell by 6.3%.23 We adjust
preference parameters ξt and γt so that the model approximates these changes in con-
sumption shares and average levels observed in the data. Similarly to the shocks to the
labor market considered in section 5.3, we assume that the shocks will last for one year
on average and diminish at rate ρ = 0.25.
Table 5 shows welfare effects from the transition incorporating preference shocks. With
preference shocks, quantifying welfare effects of the COVID-19 becomes challenging since
a new set of preference parameters directly affects welfare. Therefore, we compute wel-
fare effects from different paths of consumption before and after the COVID-19 shocks,
evaluated in terms of utility function in the initial economy. Although the level of welfare
23We approximate the effect of the COVID-19 shocks on the consumption level by a change between
the fourth quarter of 2019 and the second quarter of 2020, rather than between the first and second
quarters of 2020. We note some caution in quantifying the impact of COVID-19 on consumption from
the time series data over this short time horizon before and after the crisis. Some decline in consumption
had already begun in the latter half of the first quarter of 2020, in March in particular, and we avoid
using this quarter as a basis of comparison. Also, there was a hike in the consumption tax rate from 8%
to 10% in October 2019. The government implemented tax credits under some conditions for purchases
until June 2020, in order to alleviate negative effects on consumption caused by the tax increase and to
encourage more “cashless” transactions. Isolating pure effects of the COVID-19 crisis on consumption
from these and other factors would be a non-trivial task. For these reasons, we use a quarterly change in
consumption from 2019Q4 to 2020Q2 as approximating the COVID-19 shocks. Although the estimated
change may vary under alternative assumptions, we think the main message from the welfare comparison
across heterogeneous individuals presented in this section would remain intact.
28
-
effects requires caution in interpretation, we confirm the same pattern of heterogeneous
impact across different types of individuals, as shown in Table 5.24 Welfare effects are
more negative for females than males, contingent workers are hit harder than regular
workers and so are the low-educated than the high-skilled.
Table 5: Welfare Effects with Preference Shocks (aged 25-64, in PV-CEV)
Emp. type Education
All Regular Cont. High Low
All −1.11 −0.80 −3.49 −0.65 −1.49Male −0.96 −0.87 −3.49 −0.64 −1.31Female −1.44 −0.59 −3.49 −0.72 −1.77
5.4.2 Employment Type Definitions
As discussed in section 2.2 and in more details in appendix A, there is some discrepancy
in the definition of “contingent” workers between the MLS and the LFS or ESS. In the
LFS, the employment type is based on how workers are called by employers and it is
classified similarly in the ESS. The MLS does not have an equivalent category and we use
earnings of “part-time” workers, as representing that of contingent workers in the model.
Part-time workers are defined as those who work either fewer hours per day or fewer days
than regular workers.
Since the classification of the LFS/ESS is not based on work hours, we may include
workers categorized as “full-time” in the MLS that are grouped as contingent workers
in the LFS/ESS as well as “part-time” workers in the MLS that are grouped as regular
workers in the LFS/ESS. The fraction of the latter, i.e. the number of regular workers
in the LFS/ESS who work fewer hours as “part-time” workers in the MLS is very small
at a few percentage of all regular workers in the 2017 data. Hence, the difference in
the distribution of employment types in the two databases is mostly attributable to the
former.
Table 6 below shows fractions of regular workers in the MLS, LFS and ESS for workers
aged 15 and above. The last two columns, fractions of regular workers in the LFS and
ESS, demonstrate that the classification is close in the two.
24Although the focus of the analysis is a relative difference of welfare effects across different types of
individuals, the levels of welfare effects also differ from those in the baseline without preference shocks
since we are imposing the same pre-crisis preference in the computation. For example, shocks to the
share parameter induce more consumption of ordinary goods, which carry more weight in the pre-crisis
preference and make the welfare effects less negative (i.e. closer to zero), compared to the welfare effects
evaluated without preference shocks. Other equilibrium effects also affect the magnitude of the welfare
evaluated under the pre-crisis preference. We note, however, that since preferences are not type-specific,
these effects do not affect our relative comparison of welfare across different types of individuals.
29
-
Table 6: Fractions of Regular Workers (aged 15 and above, in %, 2017)
MLS LFS ESS
All 76.5 65.2 64.5
Male 86.3 80.1 80.0
Female 67.3 46.6 45.6
The fraction of regular workers is larger in the MLS than in the LFS/ESS. The dif-
ference is particularly large for females, implying that there are many female workers,
called contingent workers by employers, who work as many hours as regular workers. By
assuming the (more negative) earnings shocks of “part-time” workers for these workers,
we could over-estimate the shocks to regular workers in the simulations.
In order to assess such effects on wage shocks calibrated from the MLS data, we
compute wage shocks ωe,d,t assuming that part of contingent workers, corresponding to
the difference between the fractions in the MLS and the LFS shown in Table 6, are in
fact regular workers of the MLS, and include them carrying the wage shocks of regular
workers instead of contingent workers.
As a result, wage shocks of contingent workers change from 0.9748 to 0.9796 for the
ordinary sector and from 0.9757 to 0.9802 for the social sector. Since the shocks to “full-
time” workers are milder than those to “part-time” workers, the shocks of contingent
workers would become milder.
Table 7 shows welfare effects of the COVID-19 shocks under an alternative specification
of wage shocks, adjusted for the discrepancy in the distribution of employment types
between the MLS and LFS/ESS as described above. Compared to the baseline results in
Table 3, negative welfare effects on contingent workers are slightly mitigated, increasing
from −5.0% to −4.0% for both males and females, for example, although main resultsremain unchanged.
Table 7: Welfare Effects under Alternative Wage Shocks (aged 25-64, in PV-CEV)
Emp. type
All Regular Cont.
All −1.95 −1.59 −4.69Male −1.73 −1.64 −4.29Female −2.43 −1.45 −4.79
5.4.3 Duration of Shocks
In the baseline simulations, we assume that the COVID-19 shocks will diminish at rate
ρ = 0.25 on a quarterly basis and last for 4 quarters in expectation. We consider two
alternative scenarios in which shocks last for 2 and 6 quarters on average. Table 8 shows
30
-
how welfare effects vary by duration of the shocks in the labor market. Not surprisingly,
welfare loss is magnified when shocks last longer and exacerbate welfare loss of the vulner-
able more. The table shows the difference across genders, but the pattern of heterogeneous
welfare effects across other dimensions remains the same as in the baseline simulations
presented above.25
Table 8: Welfare Effects and Shock Durations (aged 25-64, in PV-CEV)
Baseline
Duration 6 months 12 months 18 months
All −1.01 −1.99 −2.94Male −0.89 −1.75 −2.58Female −1.29 −2.53 −3.72
5.4.4 Welfare Effects across Household Types
The unit of our analysis is an individual, and we do not explicitly consider a family
structure in the baseline simulations. We observed a significant difference in the labor
market experience across individuals by their characteristics. An especially large difference
was observed between regular and contingent workers.
In this section, we simulate a model to infer how a household that consists of two
earners of particular types may fare against other types of married households. We hypo-
thetically construct earnings of a typical male and female individual engaged in a regular
or contingent job. Four types of households that differ by gender and employment type
of spouses are constructed. We then quantify welfare effects of the COVID-19 shocks on
these four types of households and compare them.
Figure 15 shows the welfare effects married individuals in terms of PV-CEV, present
discounted value of consumption equivalent variation, for each individual in a two-earner
household of different combinations of spouses’ employment type. As in previous figures,
they are expressed in terms of average earnings of each type of households in the initial
economy. Not surprisingly, members of two-earner households that consist of two con-
tingent workers suffer the most. The negative effect of the COVID-19 is the smallest for
married households with two regular workers.
25We do not show all the results under alternative duration assumptions due to a space constraint, but
they are available from the authors upon request.
31
-
Figure 15: Welfare Effects of Married Individuals by Family Type (in PV-CEV)
Our model assumes exogenous labor supply and also abstracts from home production,
which potentially would be an important factor, since how the COVID-19 shocks affect
allocations of time endowment, especially within a household may have important welfare
implications. Not only an increase in unemployment, but also a rise in the amount of
remote work most likely increases total time spent at home and the demand for home
production. Moreover, the need to take care of children during the school closure period
increases a demand for hours spent at home and affects households with children and
labor supply of working parents.
The reallocation of productions in the market and at home also affects the evaluation
of effects on the aggregate economy. Leukhina and Yu (2020) examined changes in time
allocations of households during the COVID-19 recession using the American Time Use
Survey. They estimate that the monthly value of home production after the COVID-19
crisis rose by the amount equivalent to more than 10% of a decline in the monthly GDP.
Alon et al. (2020b) also argue that increased childcare needs at home contribute to a
rise in unemployment of females. Although it goes beyond the scope of the paper to fully
incorporate these factors, we note how the COVID-19 crisis affected time allocation of
individuals in different types of households is an important research theme for future.
6 Conclusion
In this paper, we document heterogeneous responses in employment and earnings to the
COVID-19 shocks during the initial months after the onset of the crisis in Japan. We
then feed these changes in the labor market into a life-cycle model and evaluate welfare
consequences of the COVID-19 shocks across heterogeneous groups of individuals.
We find that negative effects of the COVID-19 shocks in the labor market significantly
vary across people of different age group, gender, employment type, education level, in-
dustry and occupation. In each dimension, the shock is amplified for those who earn less
32
-
prior to the crisis. Contingent workers are hit harder than regular workers, younger work-
ers than older workers, females than males, workers engaged in social and non-flexible
jobs than those in ordinary and flexible jobs. Our study identifies groups of individuals
that are more severely hurt than others from the COVID-19 crisis, and suggests how the
policy could be structured, which aims to reach the most vulnerable and the most severely
affected.
Although the scope of the paper is to evaluate short-run impacts of COVID-19 in
the labor market during the initial months of the crisis, there may well be other effects
triggered by the crisis, such as structural changes in the labor market over the medium
and long-run. Such changes may also depend on how long various shocks we observe at
this moment will persist and whether they will be repeated multiple times. These topics,
which cover a longer time horizon, are left for future research.
References
Acemoglu, D., V. Chernozhukov, I. Werning, and M. D. Whinston (2020). A multi-risk
SIR model with optimally targeted lockdown. NBER Working Paper No. 27102.
Albanesi, S., R. Gihleb, J. Kim, and J. Huo (2020). Household insurance and the
macroeconomic impact of the novel Corona virus. Working Paper.
Alon, T. M., M. Doepke, J. Olmstead-Rumsey, and M. Tertilt (2020a). The impact of
COVID-19 on gender equality. NBER Working Paper No. 26947.
Alon, T. M., M. Doepke, J. Olmstead-Rumsey, and M. Tertilt (2020b). This time it’s
different: The role of women’s employment in the great lockdown. Working Paper.
Ando, M., C. Furukawa, D. Nakata, and K. Sumiya (2020). Fiscal responses to COVID-
19 crisis in Japan: The first six months. RIETI Policy Discussion Paper, 20-P-018.
Aoki, K., A. Michaelides, and K. Nikolov (2016). Household portfolios in a secular stag-
nation world: Evidence from Japan. Bank of Japan Working Paper Series, No.16-
E-4.
Coibion, O., Y. Gorodnichenko, and M. Weber (2020). Labor markets during the
COVID-19 crisis: A preliminary view. NBER Working Paper No. 27017.
Dingel, J. I. and B. Neiman (2020). How many jobs can be done at home? NBER
Working Paper No. 26948.
Esteban-Pretel, J., R. Nakajima, and R. Tanaka (2011). Are contingent jobs dead ends
or stepping stones to regular jobs? evidence from a structural estimation. Labour
Economics 18, 513–526.
Fukui, M., S. Kikuchi, and C. Goalist (2020). Job creation during the COVID-19 pan-
demic in Japan. Working Paper.
33
-
Glover, A., J. Heathcote, D. Krueger, and J.-V. Rios-Rull (2020). Health versus wealth:
On the distributional effects of controlling a pandemic. Working Paper.
Gregory, V., G. Menzio, and D. Wiczer (2020). Pandemic recession: L- or V-shaped?
Quarterly Review 4011.
Guvenen, F., F. Karahan, S. Ozkan, and J. Song (2017). Heterogeneous scarring effects
of full-year nonemployment. American Economic Review 107 (5), 369–73.
Huckfeldt, C. (2016). Understanding the scarring effect of recessions.
İmrohoroğlu, S., S. Kitao, and T. Yamada (2016). Achieving fiscal balance in Japan.
International Economic Review 57 (1), 117–154.
İmrohoroğlu, S., S. Kitao, and T. Yamada (2019). Fiscal sustainability in Japan: what
to tackle? Journal of the Economics of Ageing 14. Article 100205.
Kahn, L., F. Lange, and D. Wiczer (2020). Labor supply in the time of COVID-19.
Working Paper.
Kaplan, G., B. Moll, and G. Violante (2020a). The great lockdown and the big stimulus:
Tracing the pandemic possibility frontier for the U.S. Working Paper, Princeton
University.
Kaplan, G., B. Moll, and G. Violante (2020b). Pandemics according to HANK. Working
Paper, Princeton University.
Kawaguchi, D. and T. Murao (2014). Labor-market institutions and long-term effects
of youth unemployment. Journal of Money, Credit and Banking 46 (S2), 95–116.
Kawaguchi, K., N. Kodama, and M. Tanaka (2020). Small business under the COVID-
19 crisis: Expected short- and medium-run effects of anti-contagion and economic
policies. Working Paper, Hong Kong University of Science and Technology.
Kawata, K. (2020). Job seeker-vacancy mismatch with COVID-19. Working Paper.
Kikuchi, S., S. Kitao, and M. Mikoshiba (2020). Heterogeneous vulnerability to the
covid-19 crisis and implications for inequality in Japan. RIETI Discussion Paper,
20-E-039.
Kitao, S. and M. Mikoshiba (2020). Females, the elderly and also males: Demographic
aging and macroeconomy in Japan. Journal of the Japanese and International
Economies 56. Article 101064.
Leukhina, O. and Z. Yu (2020). Home production and leisure during the COVID-19.
St. Louis Fed Working Paper No. 2020-025A.
Miyakawa, D., K. Oikawa, and K. Ueda (2020). Firm exit during the Covid-19 pan-
demic: Evidence from Japan.
34
-
Mongey, S., L. Pilossoph, and A. Weinberg (2020). Which workers bear the burden of
social distancing policies? Working Paper.
Okubo, T. (2020). Spread of COVID-19 and telework: Evidence from Japan. Covid
Economics 32, 1–25.
Watanabe, T. and Y. Omori (2020). Online consumption during the COVID-19 crisis:
Evidence from Japan.
Yokoyama, I., K. Higa, and D. Kawaguchi (2019). Adjustments of regular and non-
regular workers to exogenous shocks: Evidence from exchange-rate fluctuation. In-
dustrial and Labor Relations Review . forthcoming.
A Data Appendix
A.1 Labor Force Survey (LFS)
Sample: The Labor Force Survey (LFS) is a cross-sectional household survey con-
ducted by the Ministry of Internal Affairs and Communications (MIC). The LFS is es-
tablished to elucidate the current state of employment and unemployment in Japan. The
survey was first conducted in July 1947. For our research propose, we use the monthly
data, known as the “Basic Tabulation,” for the period from January 2013 to June 2020.
The survey unit is a household residing in Japan, excluding foreign diplomatic and con-
sular corps, their family members, and foreign military personal and their family mem-
bers. For the “Basic Tabulation,” approximately 40 thousand households are selected.
The questions on employment status are asked to only members aged 15 years or over.
The LFS is conducted as of the last day of each month (except for December), and the
employment status is surveyed for the week ending the last day of month.26
Definition of Variables: Employment status of the population aged 15 years and
above is classified according to activity during the reference week. Our interest is the
number of employed persons among the population aged 15 years and above, especially
aged 25 to 64. Employed persons consist of the employed at work and the employed not
at work. Employed persons at work are defined as all persons who worked for (1) pay or
profit, or (2) worked as unpaid family workers for at least one hour. Thus, LFS does not
include people with jobs but not at work as employed at work. For example, those who
did not work but received or were expected to receive wages or salary are classified as an
employed person not at work.
Employed people also consist of employees, self-employed worker, and family workers
according to their main job. We use employees (those who work for wages or salaries) and
26More detailed information can be found here: https://www.stat.go.jp/english/data/roudou/
pdf/1.pdf
35
https://www.stat.go.jp/english/data/roudou/pdf/1.pdf https://www.stat.go.jp/english/data/roudou/pdf/1.pdf
-
classify them as regular or contingent (non-regular) based on what they are termed by
their employers. The regular employment type includes executives of companies or cor-
porations and regular staff who are termed “regular (seiki) employees.” The contingent
(hiseiki) employment type includes part-time workers, albeit (temporary workers), dis-
patched workers from a temporary labor agency, contract employees, entrusted employees,
and others.
Industry classification follows the basis of the Japan Standard Industrial Classification
(JSIC) according to the main types of business and industries of establishments, as revised
in October 2013. We allocate industries into two sectors, which we call ordinary and social
sectors.
Occupations are classified based on the Japan Standard Occupational Classification
(JSOC), as revised in December 2009. We allocate them into two occupations, which we
call flexible and non-flexible occupations.
Note that the samples of both industry and occupation are all workers aged 15 to 64,
including not only employees (regular and contingent workers) but also other types of
workers (self-employed worker and family workers), since more granular age and employ-
ment type categories cannot be obtained from publicly available aggregate data.
A.2 Monthly Labor Survey (MLS)
Sample: The Monthly Labor Survey (MLS) is a cross-sectional monthly survey con-
ducted by the Ministry of Health, Labour and Welfare (MHLW). The MLS is established
to measure changes in employment, earnings, and hours worked on both national and pre-
fectural levels. The survey was first conducted in July 1923. For our research propose, we
use the monthly national data for the period from January 2013 to June 2020. The MLS
was conducted on approximately 33 thousands establishments, selected from all private
and public sector establishments normally employing five or more regular employees and
belonging to 16 categorized sectors. Surveys are conducted monthly and use values as of
the end of each month.27
Definition of Variables: In this paper, we use the monthly data for contractual
cash earnings of regular employees. The regular employees are defined as workers who
satisfy condition (1) those who are employed for an indefinite period of time, or (2)
those employed for a fixed term of one month or more. Then, the regular employees are
classified as “full-time employees” and “part-time workers.” In section 5, we follow this
definition as employment type. The part-time workers are those who satisfy condition (1)
whose scheduled working hours per day are shorter than ordinary workers, or (2) whose
27More detailed information can be found here: https://www.mhlw.go.jp/english/database/
db-slms/dl/slms-01.pdf
36
https://www.mhlw.go.jp/english/database/db-slms/dl/slms-01.pdf https://www.mhlw.go.jp/english/database/db-slms/dl/slms-01.pdf
-
scheduled working hours per day are the same as ordinary workers, but whose number of
scheduled working days per week is fewer than ordinary workers.
The 16 industry categories follow the basis of the JSIC according to the main types
of business and industry of establishments, as revised in October 2013. The 16 industry
categories are a slightly less granular categorization than that of the LFS. Then we simi-
larly allocate industry into two categories, which we call ordinary and social by following
the strategy taken in Kaplan et al. (2020b).
• Ordinary Sector: mining and quarrying of stone and gravel; electricity, gas, heatsupply and water; construction; manufacturing; wholesale; transport and postal
service; information and communications; finance and insurance; real estate, goods
rental and leasing.
• Social Sector: retail trade; education and learning support; medical, health care andwelfare; living related, personal, and amusement service; accommodations, eating
and drinking places; scientific research, professional and technical services; com-
pound services; services, n.e.c.
We use as earnings contractual cash earnings plus bonuses in this paper. Cash earnings
are the amount before deducting taxes, social insurance premiums, trade union dues or
purchase price, etc. Contractual cash earnings are defined as earnings paid according to
a method and conditions previously determined by labor contract, collective agreement,
or wage regulations of establishments. The contractual cash earnings consist of scheduled
cash earnings and non-scheduled cash earnings, which are overtime pay. Overtime pay is
the wages paid for work performed outside scheduled working hours, such as at night and
in the early morning. Note that contractual cash earnings include a salary paid without
actual labor, such as leave pay.
A.3 Employment Status Survey (ESS)
Sample: The Employment Status Survey (ESS) is a cross-sectional household survey
conducted by the Ministry of Internal Affairs and Communications (MIC). The ESS aims
to obtain basic data on actual conditions of the employment structure at both national
and regional levels by surveying the usual labor force status in Japan. The ESS was
conducted every three years between 1956 and 1982, and has been conducted every five
years since 1982. For our research propose, we use the latest data collected in October
2017. The survey unit is a household of members aged 15 years and above residing
in Japan except for (1) foreign diplomatic corps or consular staff (including their suite
and their family members), (2) foreign military personnel or civilians (including their
family members), (3) persons dwelling in camps or ships of the Self-Defense Forces, (4)
persons serving sentences in prisons or detention houses, and (5) inmates of reformatory
37
-
institutions or women’s guidance homes. Approximately 490 thousand households living
in sampled units are selected.28
Definition of Variables: To obtain the distribution of employees with various char-
acteristics, we use the “order-made” data and focus on employees aged 20 and over.
For characteristics of employees, we follow the information about age, gender, education,
employment type, sector, occupation, and income.
Age is counted as of September 30, 2017. In this paper, we use data for the 10-year
age groups: 30s, 40s and 50s. Education status is defined according to the information on
the survey date. In this paper, we allocate education status into two types, which we call
high and low. We define employees as high-skilled if they have a college or higher degree,
and low skilled otherwise.
In this paper, we focus on employees and classify them into two types of employment:
regular and contingent. The regular employment type includes executives of companies or
corporations and regular staff who are termed “regular (seiki) employees.” The contingent
(hiseiki) employment type includes part-time workers, albeit (temporary workers), dis-
patched workers from a temporary labor agency, contract employees, entrusted employees,
and others.
Industry classification follows the basis of the JSIC for the main types of business and
industries of establishments, as revised in October 2013. We allocate industries into two
sectors, which we call ordinary and social sectors.29
Occupations are classified based on the JSOC, as revised in Dece